Unexplained interference in PSU test lab

[Phaedrus]

Hi guys,

I'm having some weird issues in my lab. I do a lot of ATX power supply testing, and part of this testing involves measuring output ripple and noise. The problem is, I'm getting anomalously high readings, typically 30-50mV higher than comparable labs on the exact same power supplies. The actual ripple waveform is the same magnitude, +/-5mV, when measured with V-bars, but I'm getting large spikes which throw the automated reading off, and using the V-bars is too time-consuming when I have to make 30-40 measurements.

Here is the test setup:

Tektronix TDS3014C DPO
Fluke 87V DMM
BK Precision 8512 1200W load
2x BK Precision 8540 150W loads
Extech 380820 Power Analyzer
Custom test fixture


I thought the issue was the test fixture, where the power supply cables, DC loads, and scope probes attach to the PCB. I replaced the SMA connectors for the probes, I replaced the 0.1uF ceramic caps and 10uF electrolytic caps demanded by Intel testing procedures (and upgraded from Taicon caps to equivalent NCCs), and cleaned all connectors with isopropyl. No significant change, for quite a bit of effort.

Using different probes or scope channels makes no difference. I'm considering if I might be picking up switching noise from the DC loads, or possibly conductive EMI from the mains. But I'm not sure how I could test for that. I'm kind of at a loss here.

[minime72706]

What do the spikes look like? Are they periodic and always roughly the same size? Whether it's periodic or not, how often does it seem to occur? How wide are they?

[Phaedrus]

I'll post a scope shot.

[Phaedrus]

The weird thing is, it matches up well with the peak of the waveform. But that spike shouldn't exist! We had an identical unit with the same spike, and when tested in another lab the spike wasn't there.

This unit at this load (can't give you details, sorry) is supposed to be at ~40mVpk-pk ripple on the +12V rail. V-bars on the actual waveform show ~38mV. But I've got an 80mV spike.


Scope is at 20MHz, 1M impedance, Sample mode, normal trigger, etc. Using 1x passive probes with SMA connectors.

[notsob]

Try turning your lights off

[Galaxyrise]

We had an identical unit with the same spike, and when tested in another lab the spike wasn't there.

How tough is it to move things between labs?  If you could take all your equipment to the other lab and try replacing each part of the test rig one by one you should be able to isolate it to either a piece of equipment or the lab itself. 

[Phaedrus]

By other lab I mean Europe. And lights make no difference.

[TheBay]

Sounds like some interference on mains, do you have homeplugs, energy saving light bulbs, fluorescent lighting....

Have you scoped mains power?

[Rufus]

The weird thing is, it matches up well with the peak of the waveform. But that spike shouldn't exist! We had an identical unit with the same spike, and when tested in another lab the spike wasn't there.

Lined up with ripple indicates it is from a switching transient in the PSU. If it isn't on the output then the scope is seeing some common mode signal with respect to the scope mains earth input (or other ground connections?).

Checking with a portable scope (no earth) would quickly tell you something if you have access to one.

I would look at and experiment with the mains supply to the PSU, scope, and other test equipment. Try clip on ferrite cores on the scope mains lead and possibly probe lead they shouldn't make any difference but if they do you gained a clue.

[alm]

Sounds like a common-mode signal picked up by the probe (although a coaxial SMA connection should have a very small loop area) indeed. Any chance you can cut one of the traces going to the test point and short the test point near the end of the trace? If you still see a signal with a shorted scope probe, then you definitely have a common-mode signal that may not exist in the actual circuit. Bob Johnson and Doug Smith have written on this topic, I'm sure some of it is available online.

What kind of probe is this? A common high-impedance 1x probe is unlikely to have 20 MHz of usable bandwidth at reasonable test point impedances.

[Phaedrus]

I checked with different sockets, with other devices in the building turned off, I tried power through the Power Analyzer (w/ isolation) and straight from the wall socket. Even tried a cheap ferrite bead, for all that's worth. No result. It's not on the AC side, as far as I can tell.

It's either the scope/probes, the test fixture, or interference from the DC loads.

[KJDS]

Does the burst of noise always appear at the top of the general ripple?

[AndyC_772]

Can you post a picture of your test fixture, showing exactly how the connectors, probes and any other components are physically laid out? I agree with alm, it's almost certainly a probing problem, and it can probably be fixed with changes to how the various parts of the system are connected together.

[ajc]

I had a similar issue in my lab, but turned out to be some small flourescent lamps I used above my lab desk.  Might also be worth checking / confirming you have a good earth connection in your lab. (another issue I found in my case)

Out of interest, how are you grounding your probes?  Any chance you are picking up the signal in the ground of the probe?

[notsob]

are you in an isolated building or do you have co-tenants around you
i'e the interference could be coming from above-below or in a room to the side of you.

[AndyC_772]

If the spikes are in sync with the switching waveform, then one way or the other, they're coming from the PSU.

A spike at the peak of a waveform, where the output voltage stops ramping one way and starts ramping in the other direction, is the absolutely classic manifestation of switching noise. It's either very poor PSU design (switching noise couples capacitively across the output inductor, and is not properly filtered by the PSU's output caps), or more likely, it's common mode noise which is made visible by lead inductance in the scope probe's ground.

It can't possibly be "external" interference, it's something inherent in the system of PSU + test setup + scope.

[Phaedrus]

The PSU design isn't the issue; like I said, these units have been tested in another lab and they don't have this problem. It's a high-end, high-wattage LLC resonant full bridge converter with interleaved PFC, designed by a very respectable firm. >90% efficiency at 50% load, very tight voltage regulation, etc. And very low ripple, when measured by anyone except me. 

[Phaedrus]

Could a voltage potential between scope ground and PSU ground cause this issue?

[Phaedrus]

Also worth noting that the amplitude of this spike scales fairly linearly with load on the power supply, and it exists when testing other power supplies as well. At 0W load there's no spike. At 500W load there's a ~50mV spike. At >1000W load there's an 80mV spike.

[AndyC_772]

Can you post a picture of your test fixture, showing exactly how the connectors, probes and any other components are physically laid out? I agree with alm, it's almost certainly a probing problem, and it can probably be fixed with changes to how the various parts of the system are connected together.

Just reiterating...

[tld]

Any chance the noise-difference between sites could be due to different mains-impedance?

(Asking because I'm not sure, more than anything else)

[Phaedrus]

Here you are:
http://imgur.com/48QrdB2,WzoRpwn,OsSYmTx,msSUPkI,WEo9I8m,sG6XBy3,xBq6OaQ,WUYSWWm


I did significant amounts of work on the test fixture, so forgive the flux residue. It's my #1 suspect as well (though I want to rule out other things). It had damage to the bottom, including a couple of severed traces, and a couple of damaged connectors on top.

What I did:
Removed all SMA connectors and replaced the ones we actively use
Replaced ceramic 0.1uF caps
Replaced electrolytic 10uF caps
Repaired damaged traces with tinned copper wire
Added tinned copper wire and solder to some high-current traces to increase thermal mass and reduce impedance
Put board on standoffs


I can't see anything I did adding noise like this.

[KJDS]

If you double the length of those SMA leads, then I'll put money that the length of ringing on that sharp pulse will double.

Have a close measure of the ringing frequency. Work out what length lead that would correspond to and then look at your current lead lengths.

Please note the above is conjecture, not fact, but it's fairly educated conjecture.

[AndyC_772]

Remove the electrolytics on the back of the test PCB and replace them with leadless ceramic caps. Ripple and noise on a PSU are almost always measured across a decoupling cap, rather that just at the end of a long wire as you have now.

A small (say, 100nF) ceramic cap won't have any effect on the magnitude of the 'real' ripple, but it will help suppress the unwanted high frequency switching spikes. If you're concerned about affecting the measurement accuracy, bear in mind that any PC motherboard will have many ceramic caps on it anyway, so you're actually testing a scenario which is more representative of a real use case, not less.

There's an argument that in fact there's little point in having effective high frequency decoupling in a PSU that's designed to be connected to its load over long wires. The only really good reason is EMC. If a load requires low noise at high frequencies, it will always need to have effective, local decoupling anyway, and a decoupling cap at the opposite end of a power cable is useless.

[fcb]

Are you testing at 110V and the EU lab at 230V?